Introduction to Multiple Target Tracking
简单介绍了多目标物跟踪的流程、方法。
1、single target tracking (STT) 单目标跟踪,不需要association和assignment。
2、 multiple target tracking (MTT) 多目标跟踪,很难,原因:
①、Target or detection distribution,目标分布很紧,不同目标、轨迹紧挨着。
②、Probability of detection (Pd) of the sensor,识别的概率比较低,容易丢帧,或者错误匹配。
③、Sensor resolution,传感器分辨率低。
④、Clutter or false alarm rate of the sensor,误识别。
⑤、The number of targets and detections,轨迹或者目标多,计算指数增长。
3、工作流程,略
4、Detection,识别
传感器对于一个target(真实目标),只有一个detection(探测)。当然也可以有0个探测,或者虚假目标(Farse alarm)。
某些传感器对于一个target可能有多个detection,但是对于这些算法,必须先聚类。(such as trackerGNN, trackerJPDA, and trackerTOMHT)。
5、Gating and Assignment
简单介绍了GNN、JPDA、MHT方法。
6、Track Maintenance
Track maintenance refers to the function of track initiation, confirmation, and deletion.包括三块。
7、Filtering
①、用track预测
②、匹配
③、更新track
8、Tracking Metrics ???
评价目标跟踪算法,有三类:
① 、trackAssignmentMetrics :评价目标匹配。 provides indexes such as number of the track swaps, number of divergence steps, and number of redundant assignments.
②、 trackErrorMetrics :算track误差的。
③、 trackOSPAMetric:OSPA,provides three scalar error components — localization error, labelling error, and cardinality(基数?) error to evaluate tracking performance.
9、Non-Assignment-Based Trackers???
这部分也看不懂。
trackerGNN
, trackerJPDA
, and trackerTOMHT
are assignment-based trackers, meaning that the track-to-detection assignment is required. The toolbox also offers a random finite set (RFS) based tracker, trackerPHD. You can use its supporting features ggiwphd to track extended objects and gmphd to track both extended objects and point targets.